Symantec traps autonomous AI attack agent in honeypot
Briefing by News4U · First published
The Register reports Symantec researchers captured an AI agent built to hijack exposed models for up to 100,000 unattended steps of scanning, exploitation and crypto mining.
The Register reports that Symantec Threat Hunter Team researchers trapped an autonomous AI attack agent in an AI honeypot over the summer, revealing instructions that let the agent scan the internet for exposed OpenAI-style endpoints and hijack them for up to 100,000 unattended steps. [1]
The agent, built with layered jailbreaks and 74 offensive tool definitions, was designed to exploit common web frameworks, steal databases and credentials, execute remote code and deploy cryptocurrency miners in what its creators called “Skynet mode.” [1]
Sources
- Autonomous attack agent in 'Skynet mode' hijacks other models to carry out its evil deeds
The Register · · Excerpt